https://github.com/pyronear/smoke-localization

https://github.com/pyronear/smoke-localization

Last synced: about 1 year ago
JSON representation

• Host: GitHub
• URL: https://github.com/pyronear/smoke-localization
• Owner: pyronear
• License: apache-2.0
• Created: 2024-02-01T09:37:36.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2025-04-16T12:54:31.000Z (about 1 year ago)
• Last Synced: 2025-05-08T16:28:44.785Z (about 1 year ago)
• Language: Jupyter Notebook
• Size: 177 MB
• Stars: 7
• Watchers: 6
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Smoke localization

As soon as a suspicious smoke is detected, there is the need to precisely geolocate it, so the fire services can intervene quickly. The following tools aims to estimate a given point location from an image. 

## Installation

1. Install local packages and dependancies (from this directory):

 -  `pip install horizon/` 

 - `pip install georefcam/` 

2. Test the notebooks in `demo_notebooks/`.  

## Pipeline

![](plots/pipeline_white.svg)

1. Load Digital Elevation Model (DEM) from image location

![](plots/open3d_dem.png)

2. Extract 360° skyline from viewpoint

![](plots/viewpoint_skyline.png)

3. Extract skyline from image (with Depth-Anything and filters)

![](plots/image_skyline.png)

4. Compare skylines and find best alignment, this corresponds to the camera azimuth.

![](plots/skylines_comparison.png)

5. Place a virtual camera in DEM at correct azimuth and generate depth map.

6. Estimate depths from images (Depth-Anything)

7. Correct the virtual camera by comparing the depths

![](plots/depth_correction.png)

8. Get the location of any point in the image